Condition Monitoring: Vibration, Thermography, and Oil Analysis

Condition monitoring means measuring how a machine is doing while it runs, so you can repair it when it needs repair and not before. Scheduled maintenance replaces parts by the calendar. Condition monitoring replaces them when the data says they are wearing out. This is also called predictive maintenance. It relies on one fact: most machines give warning signs long before they stop.

Four technologies do most of the work: vibration, infrared thermography, oil analysis, and ultrasound. You do not need to be a certified analyst to help. Technicians collect the data, spot changes, and act on what the reports say. Reading about these methods does not make you an analyst, though. Vibration and thermography analysis have their own training and certification schemes, such as those built on ISO 18436.

What You Will Be Able to Do

• Explain how early detection gives you time to plan a repair.
• Take vibration readings the same way every time and match common patterns to faults.
• Read a thermal image by comparing like with like, and stay out of live panels.
• Pull a clean oil sample and read the main lines of a lab report.
• Turn a finding into the right action.

How Early Warning Works

Failure rarely happens all at once. A bearing develops a tiny defect. Weeks or months later it is noisy. Then it runs hot. Then it seizes. The earlier you detect it, the longer you have to plan. Reliability engineers call this window the P-F interval, the time between the point P where a problem can first be detected and the point F where it fails.

Each technology sees a different stage. Ultrasound and vibration usually catch a bearing problem first. Oil analysis picks up wear metals as they build. Heat tends to come later. By the time you can hear it or feel it by hand, you are near the end of the window. That is why plants use more than one method, and why a monthly reading on a part that fails in two weeks is no use. Your check interval must be shorter than the P-F interval.

Baselines and Trending

A single reading tells you little. The trend tells you what is happening.

1. Take a baseline when the machine is known to be healthy, such as after installation or an overhaul.
2. Take each later reading at the same point, in the same way, under similar speed and load.
3. Record the speed, load, and temperature with each reading.
4. Plot the readings over time.
5. Compare them with alarm limits set by your program, the machine maker, or a standard.

Worked example, for overall vibration on a fan bearing in inches per second:

Each number might sit under a fixed alarm limit, but the trend shows a reading nearly four times the baseline, and the rise is speeding up. Many programs review any machine whose reading doubles from its baseline. This one should have been looked at in April. Shorten the check interval and order the parts now.

Vibration Analysis

Vibration analysis is the most widely used predictive technology. Every rotating machine vibrates. A problem changes the pattern in ways that point to the cause.

Readings show how hard the machine is shaking (the amplitude) and how often (the frequency). To use the frequency, you need the running speed. One revolution per second is 1 hertz (Hz), and running speed is called 1x. A shaft at 1,800 RPM turns 1,800 ÷ 60 = 30 times per second, so 1x is 30 Hz, 2x is 60 Hz, and 3x is 90 Hz.

Gear mesh frequency is the number of teeth times the shaft speed. A 24-tooth gear on the 30 Hz shaft meshes at 24 x 30 = 720 Hz. Peaks beside that number, called sidebands, point to a worn or damaged tooth or an off-center gear.

Bearing defects usually show first in very high frequencies. As the damage grows, peaks appear at the bearing's own fault frequencies, which depend on its geometry. Late in its life the whole spectrum gets noisy. Catching this early can give you months of notice.

Taking Vibration Readings

1. Mount the sensor, called an accelerometer, on the bearing housing, as close to the bearing as you can. Do not use a guard, cover, or sheet metal, which shake on their own.
2. Measure in three directions: horizontal, vertical, and axial, which is along the shaft.
3. Mark each measurement point with a paint pen or a fixed mounting pad, so the next reading is taken in the same spot.
4. Let the machine reach its normal speed and load before you start.
5. Record the conditions and upload the data.

Warning: You are taking these readings on a running machine. Keep guards on and stay clear of rotating parts. Tie back loose clothing and hair. If a point cannot be reached with guards in place, have a permanent sensor or a mounting point installed outside the guard. Never remove a guard for a reading.

Severity Levels

Standards such as ISO 20816, which is replacing the older ISO 10816 series, sort overall vibration into zones. The limits depend on the machine's type, size, and mounting, so look them up for your machine and do not copy them from another one.

Infrared Thermography

An infrared camera turns heat into a picture. It shows where something is hotter than it should be, without touching it. Common targets are:

• Electrical connections. A loose or corroded connection adds resistance and heat, which can lead to failure or fire.
• Motors and bearings. A hot spot on a housing can mean a lubrication or bearing problem.
• Couplings. Misalignment makes heat at the coupling.
• Steam traps. A trap that has failed open shows the downstream pipe running too hot.
• Fuses and breakers. One phase hotter than the others points to a bad connection or unbalanced load.

Thermography works by comparison. An absolute temperature means little by itself, so compare the same part on all three phases, or identical machines under similar load. If three identical motors on one line run at 140, 142, and 172°F under the same load, the third one has a developing mechanical or electrical problem. It needs investigating even though it is still running. Check for a bearing problem, blocked cooling fins, overload, or an electrical imbalance.

For electrical parts, ANSI/NETA testing guidance gives this starting point when comparing similar components under similar load. Your plant's criteria take priority.

Three things skew a scan:

• Load. Heat rises with the square of the current, so a lightly loaded circuit can hide a bad joint. Electrical surveys are usually done with equipment at about 40 percent of its rated load or higher.
• Emissivity. Shiny metal reflects heat and reads falsely cool. The camera needs the right setting, or you can aim at a dull surface nearby.
• Reflections. Sun, lights, and even your own body can show up as false hot spots. Move and see if the spot moves too.

Warning: Do not open an electrical panel to scan it unless you are a qualified person wearing the arc-rated PPE your site requires. Opening a live panel exposes you to shock and arc flash. Where you can, use infrared windows, which let you scan with covers closed.

Electrical panels are commonly scanned at least once a year, and rotating equipment monthly or quarterly. The 2026 edition of NFPA 70B also recognizes permanently installed temperature monitoring. Compare each scan with baseline images taken when the equipment was new or just serviced.

Oil Analysis

Oil analysis is like a blood test for a machine. A lab checks a sample for wear, contamination, and the health of the oil.

Pulling a Good Sample

A poor sample gives a poor answer.

1. Sample from the same spot each time, ideally a dedicated sample port in a live, flowing zone. Do not sample from the drain plug, where sludge and water settle.
2. Sample while the machine is running or within minutes of shutdown, so that particles are still mixed in.
3. Use a clean bottle and new tubing, and flush the port first.
4. Label the bottle with the machine, date, oil type, and run hours.
5. Send it right away.

Warning: Oil from a running machine can be hot and under pressure. Use the sample port as designed, and follow your procedure for the machine.

Reading a Lab Report

Elevated iron and copper together are the clearest sign of abnormal wear inside the machine. Water limits depend on the oil and the machine. Many oils can only hold a few hundred parts per million dissolved before free water appears, so use the limit your lab or the equipment maker sets. Particle counts often come as an ISO 4406 code, such as 18/16/13. Each number stands for a range of particles at a given size, and a one-step rise means roughly double the particles.

Worked example, for a gearbox sampled every quarter:

The first two samples show normal wear. Sample 3 shows a jump in both metals, and sample 4 confirms it is getting worse. Check for a vibration change, inspect the gears and bushings at the next planned stop, and sample more often until then.

Ultrasound in Brief

Ultrasound instruments pick up high-frequency sound above the range of human hearing and convert it to something you can hear and measure. They are useful for:

• Compressed air and gas leaks, which hiss in the ultrasonic range even in a noisy plant.
• Early bearing wear and lack of lubrication. Friction raises the level, and when you add grease the level drops, so you can stop when it stops falling.
• Valves that leak through when closed, and steam traps.
• Electrical tracking, arcing, and corona, by qualified people.

Use a baseline and trend it, the same way you do with vibration.

Acting on Findings

Data is only useful when someone acts on it.

1. Confirm it. Rule out a bad reading. Check the sensor mounting, the load, and the sample. Confirm with a second method if you can, such as oil analysis to back up a vibration trend.
2. Rate it. How serious is it, how fast is it changing, and how critical is the machine?
3. Write it up. Create a work order that states what was found, where, the readings, and the likely cause.
4. Plan it. Order the parts, schedule labor, and fit the job into planned downtime.
5. Check the repair. Take new readings afterward. This confirms the fix worked and gives you a new baseline.

Common Mistakes

Check Your Understanding

1. A pump runs at 3,600 RPM. Vibration shows a strong peak at 60 Hz in the radial direction, and it has climbed slowly since a mechanical seal job. What does this suggest?

3,600 ÷ 60 = 60 Hz, so this is 1x, which points to imbalance. Check for something left out of balance by the work, such as a damaged impeller or buildup, and keep trending it.

2. A thermal scan shows one lug on a breaker 22°C hotter than the matching lugs under the same load. What should happen?

That is over 15°C compared with similar parts, so it is a major discrepancy. Report it at once. A qualified electrician should make it safe and repair it. Do not open the panel yourself.

3. Oil samples from a gearbox show silicon rising, and iron starting to follow. What is the likely story?

Dirt is getting in, probably through a bad seal or breather, and it is starting to wear the gears and bearings. Fix the entry point first, then deal with the oil.

4. You replace a bearing based on vibration data. Why take another reading afterward?

To confirm the fault is gone and to set a new baseline. If the reading is still high, the cause, such as misalignment, may still be there.

Key Takeaways

• Condition monitoring detects problems early enough that repairs can be planned.
• The trend against a healthy baseline matters more than any one reading.
• Vibration patterns point to the fault: 1x for imbalance, 2x for misalignment, many multiples for looseness.
• Thermography compares like with like, and it needs load. Live panels are for qualified people only.
• Sample oil from a consistent live point. Elevated iron and copper mean abnormal wear.
• Confirm, rate, write up, plan, and verify every finding.